| Description |
Evolution of computerized hospital information systems designed to fulfill the needs of health care providers has kindled a controversy concerning how information should be represented-freetext data or precoded data. The primary advantage of freetext storage schemes (methods utilizing the natural language of care providers to store medical observations) is that they place no restrictions on information content or structure. Many health care professionals prefer this storage method, as all clinical data and correlations between data are preserved. Precoded data, in contrast, are required to realize the benefits of intelligent computerized monitoring and processing including decision support and decision-making activities. Natural language processing applications capable of identifying, extracting, and encoding information contained in freetext data provide one resolution to this controversy. Development and evaluation of such a system, designed to encode freetext admission diagnoses, were undertaken. A description of the strategies employed forms the body of this thesis. A general purpose natural language understanding system is currently under development by Dr. Peter J. Haug et al., at LDS Hospital in Salt Lake City, Utah. This system incorporates syntactic and semantic analysis techniques providing the needed foundation for the completion of the admission diagnoses coding application created during this study. Four data structures were designed to store knowledge of how to parse and encode freetext diagnoses: an event definition data model, a Bayesian network, a knowledge base, and a synonyms file. Evaluation results demonstrate the overall performance of this application to be slightly above average, accurately encoding approximately 76% of freetext admission diagnoses. Inefficiencies are primarily due to the inability of this system to generate encodings in roughly 15% of test cases. When encodings are produced, however, accuracy equals that of the current manual coding method, and with further modification, this application can partially automate the coding process. Development of this parsing application was a time-intensive undertaking requiring the cooperation, talents, an ingenuity of several dedicated inThe rewardsls. Tewards of these combined efforts may well evolve into Continued effort to refine and enhance the performance of this application can result in a clinically useful tool streamlining work flow, reducing variability in coded diagnoses, and improving overall quality. |
